Abstract

Multi-disciplinary Design Optimization (MDO) problems are widely present in industrial design practice and in the Aerospace industry, where the performances coming from the simulation of multiple disciplines have to be optimized simultaneously. Specifically in the preliminary concept design, a typical scenario for the MDO problems is represented by interwoven systems, for which some variables are simultaneously input and output parameters of different disciplines. To solve these problems, many approaches exist in literature, most of them based on the application of gradient-based optimization algorithms with the evaluation of multiple loops to get an equilibrium solution for each design. A possible limitation of this approach is generally related to the large computational cost required to obtain the equilibrium point for each design. In addition, not every input variable combination allows to get an equilibrium point, causing convergence problems to the algorithms. In this paper we propose a different methodology, based on the application of Game Theory strategies, to solve these problems by a reduced number of discipline evaluations. The idea is to decouple the main objective(s) of the optimization problem from the one related to the reach of an equilibrium point of the system. In one version of the methodology, following the definition of a Hierarchical Strategy Game, the leader player optimizes the main objectives of the problem, while the follower player is in charge of the equilibrium point objective, which is optimized for every design proposed by the leader. In the second version of the methodology, following the definition of a Competitive Strategy Game, the two objectives are solved simultaneously by two players, who act following their own criteria and exchanging information at each step. The methodology is applied to a conceptual supersonic aircraft design MDO problem, using numerical models developed in the framework of the European Project HISAC [https://cordis.europa.eu/project/id/516132]. The advantage in terms of results and computational effort, compared to a classic methodology based on equilibrium-loops, is proven in the paper.